Skip to content
Minnow
Source
Download
CHANGELOG

Releases

Every Minnow release, newest first. Preview builds are marked separately. Downloads use the stable channel when available. Follow the RSS feed.

v0.1.14-beta.20261018.38.1 PREVIEW

Minnow 0.1.14-beta.20261018.38.1 (pre-release)

Automated beta build from commit 24ea056f098c13ec0f6d87012f3360548ead39c2.

What’s Changed

Full Changelog: https://github.com/HenriGrimm/Minnow/compare/v0.1.14-beta.20261016.36.1…v0.1.14-beta.20261018.38.1

v0.1.13

Minnow 0.1.13

Minnow 0.1.13 adds pinned chats and live session updates, editable Markdown previews, clearer provider cost guidance, and new controls for MTPLX and agent CLI usage. It also brings Cursor account usage and ACP compatibility, plus a release-note writer that can summarize a draft’s changes from its commit history.

Highlights

  • Keep important chats close. Pin chats in the sidebar and keep session and issue state current across windows with delta-based updates.
  • Edit Markdown in place. Preview Markdown files in Code, make changes in the preview, and save them back to the file.
  • See provider costs before choosing a model. The manual now explains provider pricing and trade-offs.
  • Tune MTPLX models. Choose performance presets and adjust batching, scheduling, memory, and load options with settings that map to the server launch arguments.
  • Check CLI account usage. View Cursor quota and billing-cycle usage in Models. Cursor ACP handling now validates model variants and MCP identity more accurately and preserves session state through retries and restarts.
  • Draft release notes from history. The Releases editor can read the commits since the last published release and generate notes with the configured utility model.

All changes since 0.1.12

Chats, sessions, and Markdown

  • Added sidebar chat pinning so frequently used conversations stay easy to reach.
  • Added delta-based session and issue polling to reduce full reloads and prevent stale cross-window merges.
  • Added editable Markdown preview blocks in Code, with changes saved to the source file.
  • Improved chat composer resizing and skill highlighting, plus the secondary file-viewer slot used by previews.

Models and CLI integrations

  • Added MTPLX performance presets and aligned scheduling, batching, memory, and load settings with the server launch arguments. Updated the model inspector labels and validation to reflect saved values.
  • Added Cursor CLI account-usage support using its CLI authentication and dashboard APIs. Billing-cycle windows are normalized in the shared usage model, and Cursor appears in the Models and CLI quota interfaces.
  • Improved Cursor ACP model-variant and MCP identity validation, preflight retry handling, and preservation of configuration and session state across restarts.
  • Added self-update flows for installed agent CLIs.
  • Updated Claude CLI usage normalization and the shared agent CLI usage display.

Providers and release workflow

  • Added a provider costs and pricing trade-offs guide and updated model connection guidance.
  • Added a release-note writer to the Releases editor. It reads the editable draft’s target commit and the history since the last published release (or a chosen base tag), then uses a utility model to generate notes. Generated text is previewed before applying, can be undone, and requires an explicit save.
  • Updated the shipped manual, architecture context, product wiki, and regression coverage for these changes.

Full Changelog: https://github.com/HenriGrimm/Minnow/compare/v0.1.12…v0.1.13

v0.1.12

Minnow 0.1.12

Minnow 0.1.12 adds Actions and Releases to Source Control, MTPLX model hosting on Apple Silicon, and scheduled automation for GitHub issues. It also makes first-run setup more complete, adds cloning from remote repositories and GitHub plugin installation, and improves planning, file reviews, search, and runtime reliability.

Highlights

  • Run workflows and manage releases inside Minnow. Inspect GitHub runs, dispatch workflows, run local commands, and manage release notes and assets from Source Control.
  • Host models with MTPLX. Choose an engine per model, configure MTPLX load settings, and follow runtime activity in Local Server.
  • Turn labeled GitHub issues into reviewed work. Scheduler can plan and implement issues labeled minnow, test the result, and open a pull request for your review.
  • Start with the tools you use. Onboarding supports CLI agents, shared GitHub sign-in, existing GGUF models, web search setup, and clearer optional-component installation.
  • Choose how a plan runs. Create a sequential Build plan for one chat or an Orchestrate plan for a board, and pause and resume live board work.

All changes since 0.1.11

Actions and Releases

  • Added Actions and Releases sections to Source Control, with responsive lists and inspectors. Actions opens on Runs and includes Workflows and Commands.
  • Added GitHub workflow inspection, run details and logs, workflow dispatch with ref-specific inputs, and run cancellation.
  • Added saved local commands from .minnow/actions.json and package scripts. Local GitHub workflows can run through user-installed act and Docker for supported Ubuntu container jobs.
  • Local runs record workspace, branch, commit, and dirty state, retain bounded logs and metadata for seven days, and protect active checkouts from removal. A dedicated worker owns process cleanup, including workflow containers.
  • Local action secrets are encrypted and included only in credential backups; run logs are excluded from backups.
  • Added release creation, editing, publication, deletion, and asset management. Releases selects the newest loaded release automatically and renders sanitized Markdown notes, with editing separate from the overview and assets.
  • Run summaries appear immediately while jobs load. Recently viewed details are cached and refreshed in the background; completed steps start collapsed, and expanded steps and opened logs survive polling.
  • Polling preserves unsaved command and release edits. Fixed the GitHub repository-root URL used for release permission checks.
  • Added Use for releases to eligible GitHub workflows and a Releases shortcut for dispatching the chosen workflow. The mapping is shared across worktrees for the same repository; the form validates refs and inputs and rejects stale submissions.
  • Added seven agent tools: action_inspect, release_inspect, action_run, action_cancel, action_command, release_manage, and release_asset. Inspection is enabled by default; mutations ask for approval and remain blocked in Plan mode.

Local models and MTPLX

  • Added MTPLX hosting for MLX weights on Apple Silicon with macOS 14 or newer, including installed-runtime detection, cache scanning, diagnostics, and Discover downloads targeting the MTPLX cache. MTPLX is installed separately.
  • Added per-model engine selection. Validated MTPLX models can use MTPLX acceleration or the MLX engine without MTP acceleration.
  • Added descriptor-aware controls for context, MTP depth, KV quantization, reasoning, scheduling, caching, sampling, and system settings. Out-of-range saved values are adjusted with a warning, and inherited defaults are visible in the controls.
  • Added connection to matching external MTPLX daemons; eject disconnects without terminating an external daemon. Minnow-owned processes support idle timeout and one automatic restart after a healthy daemon crashes.
  • Local Server shows MTPLX prefill/decode speed, time to first token, cached tokens, MTP depth, and active requests. Fixed its live chat token counter and lifecycle/heartbeat handling.
  • Recognized vision-capable MTPLX checkpoints while avoiding false image support when those weights use the text-only MLX backend. Fixed model binding precedence for chat, library, and headless use.
  • Refined My Models and Local Server layouts, model inspectors, connection details, and load-settings popovers. Basic and Advanced controls retain edits, save automatically for the next load, and offer Use defaults.
  • Added confirmed deletion for scanned GGUF variants and split shards or MLX model folders. Loaded models and unfinished downloads are protected, and deletion refreshes the library and shared pickers.
  • Improved llama.cpp shutdown using its graceful stdin exit protocol before process termination. This is a shutdown mitigation, not a confirmed resolution of system-wide native hangs.
  • Runtime logs now replay and follow the complete recorded output in bounded chunks, preserve UTF-8 across chunk boundaries, and append lines without rebuilding completed rows. Copy includes received output; Clear view resets the visible buffer.
  • Improved CUDA toolkit discovery and build setup for local runtime forks. Added MTPLX settings to the settings backup catalog.

Onboarding, GitHub, and projects

  • Added a CLI agent setup path with detection, authentication verification, enablement, advanced settings, and default-model selection. Install and sign-in commands are shown for an external terminal during onboarding.
  • Added shared GitHub sign-in to onboarding and Settings.
  • Added a dedicated Web search step recommending Tavily, with account and API-key guidance. New search configurations default to Tavily and leave SearXNG auto-start off; existing explicit settings are preserved.
  • Removed the legacy Apps step from the wizard while retaining saved-step compatibility. The app reveal waits for first-run decisions.
  • Added managed-storage setup for loading existing GGUF models, and improved Extras installation with progress, logs, clearer messages, and retry behavior.
  • Added Create From Remote to the workspace picker and welcome screen for HTTPS and SSH repository URLs, with an editable suggested folder name and the configured project location.
  • Clone operations use existing Git credentials or SSH keys, reject existing destination folders, clean up newly created folders after failure, and retain successful clones if opening needs a retry.
  • Reworked development-server add/edit forms into anchored popovers with inline validation, free-port suggestions, and advanced fields.

Planning, boards, and GitHub issue automation

  • Planner and Plan mode now ask whether a new plan should run as Build in one chat or Orchestrate on a board. Build plans use sequential Changes/Verify steps and checklists; revisions preserve the chosen format.
  • Build previews omit board actions, and board intake rejects Build documents. Agent validation, board creation, and resync now share structure and workspace-aware dependency checks for consistent diagnostics.
  • Added Pause/Resume for live board chats, retaining attempts and worktrees and excluding paused time from attempt time limits. Work already in flight may finish; Stop retains its abort behavior. Paused boards remain paused after restart.
  • Added Scheduler’s GitHub issue watcher. It processes open issues labeled minnow, one at a time per repository, plans in a separate worktree, implements through a Builder/Tester board, and opens a PR after integration and final tests pass.
  • Watchers publish progress and failures to issue comments, show claimed issues and board/PR links, and provide Retry and Run now. Claims survive restarts to prevent duplicate processing, and publication retries discover existing PRs.
  • Removing the label or closing an active issue stops its board on the next poll. Disabling the watcher pauses polling and publication; PRs remain open for review and merge.
  • Bounded long board and worktree storage names with stable hash suffixes, enabled Git long-path handling for managed Windows operations, and added recoverable errors for invalid or incomplete checkouts. Existing board IDs and journaled paths are preserved.
  • Plugins can now install from public GitHub repository or folder URLs as well as local folders, including folders outside the active workspace. GitHub imports are pinned to the reviewed commit.
  • GitHub plugin imports download one source archive per commit to avoid per-file REST rate limits. Repository URLs can discover a single plugin in a subfolder; compressed/unpacked size limits and rejection of links and special files remain enforced.
  • Source Control changed-file reviews open the full file in the adjacent editor, highlight additions, and show removed lines inline. Staged reviews compare HEAD with the index; unstaged reviews compare the index with the working tree.
  • Fixed staged rename comparisons to use the original HEAD path and report files missing from the working tree as deleted. Commit reviews retain their side-by-side layout.
  • Added the read-only load_skill tool for discovering skills and reading their instructions and bundled references, including board and headless agents, with pagination and path guards.
  • Added the built-in frontend-design skill, moved its shared design reference into the skill package, and enabled model invocation for additional built-in skills.
  • General-mode prompts now direct agents to create requested documents directly.
  • Added Tavily Map and batch Extract tools, richer search controls and excerpts, and Search Settings credit-usage reporting with refresh and error handling.

Chat, Issues, companion sync, and reliability

  • Fixed Codex CLI vision handling across image attachments, typed history, saved-thread resumes, and dynamic tool results, while retaining text-only model restrictions and image validation.
  • Fixed concurrent Codex CLI inference and authentication handling. Improved Cursor streaming to avoid duplicate flushes and refined unread/reply bookkeeping.
  • Sub-agent cards now use accurate transcript tool totals, preserve counts when transcripts are capped, and retain completed-run duration after reload.
  • Issues show background expansion activity in lists, boards, detail views, and Code’s sidebar, including across windows. Duplicate expansion starts are ignored.
  • Linked issues and Chats rows show running, finished, failed, stopped, or interrupted chat-turn activity across windows and observer reloads. Completing a chat does not automatically resolve its issue.
  • Fixed Issues context menus closing while chats scroll. Question keyboard commands now apply only when their panel has focus.
  • Preserved focus during background editor and preview updates, and kept memory-saved cards above native browser previews.
  • Improved companion pairing with authenticated installation, device metadata hydration, and live session refresh across windows while preserving local edits and active-chat selection. Improved touch/scroll follow behavior and provider-registry retries.
  • Reduced expensive issue-state merge comparisons during refresh and save reconciliation.
  • Git status lists and diffs retain their DOM when unchanged, reducing polling and Windows accessibility churn. Fresh diffs are still fetched, and stale responses are rejected.
  • Increased the retained-generation memory budget to 512 MiB and released completed request bodies. Slow subscribers replay retained chunks lazily instead of copying them into live queues; terminal events and active replays survive memory-pressure eviction.
  • Development launches now capture server and Electron stdout/stderr, lifecycle events, exit codes, signals, and spawn failures in rotating logs under ~/.minnow/logs/runtime, including detached Electron launches.
  • Updated shipped manuals, technical references, product wiki entries, package/runtime validation, regression coverage, and bundle accounting. Added runtime lifecycle and serializer diagnostics.

Downloads and updates

Packages are included for Windows, Linux x86_64, and Apple Silicon macOS. Installed users can check for published releases under Settings → About & troubleshooting → Updates.

Full commit comparison

v0.1.12-beta.20261008.28.1 PREVIEW

Minnow 0.1.12-beta.20261008.28.1 (pre-release)

Automated beta build from commit 9648dfd10f7e0ce7046fe831a1b02e73997a4c15.

What’s Changed

Full Changelog: https://github.com/HenriGrimm/Minnow/compare/v0.1.12-beta.20261007.27.1…v0.1.12-beta.20261008.28.1

v0.1.11

Minnow 0.1.11

Minnow 0.1.11 brings issue tracking into the Code files sidebar, makes sub-agent activity easier to follow, and adds a configurable default folder for new projects. It also introduces interactive Claude sessions behind normal chat, improves Codex reasoning and context display, and fixes issue handoffs, automatic GitHub sync, literal code search, and large Brain graphs.

Highlights

  • Work on issues without leaving Code. The files toolbar now has an Issues pane with quick capture, property filters, and the same editable details and actions as the Issues app.
  • Follow delegated work at a glance. Each sub-agent appears as one timeline step with its brief, status, tool count, duration, and current activity or result.
  • Choose where projects live. Set a default project folder in Settings, and see that location before creating a project.
  • Keep Claude work moving. Interactive sessions now power chats, board agents, and enabled helper tasks, with improved startup handling, tool handoffs, and usage accounting.
  • See Codex context accurately. Reasoning output renders more completely, and the context ring uses native context occupancy rather than per-request billing totals.

All changes since 0.1.10

Issues inside Code

  • Added an Issues ticket button to the files toolbar. It switches the sidebar to the current workspace’s issues, grouped by status, with Open, All, and Closed choices and title or ID search. Browsing a worktree retains the Code workspace’s issue scope.
  • Added property filters for type, status, priority, and project, including No project. Filters combine with search and open/closed selection, appear as removable chips, and retain separate selections per workspace while the app is open.
  • Added quick capture with Create or Ctrl/Cmd+Enter: the first line becomes the title and subsequent lines become the description. Drafts survive pane switching and reloads; failed-save retries avoid creating a duplicate issue.
  • Expand in sidebar quick capture creates the issue, expands it in the background, then clears and refocuses the input so several issues can be recorded in a row.
  • Reused the full Issues detail view inside the sidebar, including editable properties, labels, description, links, workflow actions, related issues, and comments. Back restores list filters, scroll, and focus; related issues open in the same pane.
  • Unified row actions across Code and the Issues app, including viewing, editing in Issues, Send to chat, property changes, expansion, sub-issues, copying, deletion, and opening linked GitHub issues. Context actions are available through right-click, Shift+F10, or the row menu.
  • Added Filters and New issue to the blank-space context menu in the Issues list and board, plus New issue in row menus. New issue opens an empty form without copying the clicked issue or column.
  • Kept GitHub conflict controls registered independently for each open detail surface, so closing the sidebar detail does not remove conflict handling from the Issues app.

GitHub sync and issue handoffs

  • With Two-way mirror → Sync automatically enabled, newly created local issues now create their GitHub counterpart without waiting for a later edit. This includes issues added by agents or received from another window, plus synced-field changes discovered during storage merges.
  • Automatic pushes remain scoped to the current workspace. Scratch, other workspaces, initial loads, and recovery loads do not trigger a backfill of older unlinked issues.
  • Fixed issue handoffs placing their internal submitted prompt into the composer as an unsent draft. Handoffs now send directly to the created chat, with code references supplied as deduplicated attachments.
  • Kept handoff delivery bound to the original chat when you switch chats during worktree setup. Ordinary editable chat seeds still appear promptly in the composer.
  • Fixed the issue run-target picker opening in the wrong place when the sidebar rerenders after a mode selection; it stays anchored to the original Send button position.
  • Issue tickets in chat now use the available message width.

Projects and workspace controls

  • Added Settings → General → New projects → Default project folder. Choose an existing absolute folder, or clear it to restore your home folder’s Projects directory. Existing projects are not moved, and the creation form retains its per-project location override.
  • The Choose workspace page now displays the effective default location with a Change default folder link before project creation.
  • Browse… saves a selected folder immediately. Typed paths save on blur or Enter, Escape cancels the edit, and save progress and errors appear in an accessible status line.
  • Fixed Issues and Preview toolbar buttons appearing active while their pane is collapsed, and kept button state in sync with mobile sidebar opening and closing. Updated Issues and capture controls to use a ticket icon.

Claude and Codex

  • Claude Code now runs an interactive terminal session behind normal chat, requiring Claude Code 2.1.289 or newer. Text and images travel through a private connection while Minnow retains tool execution and approval handling.
  • Extended interactive Claude sessions to board agents and enabled utility roles, including helpers without a chat ID. One-shot helpers receive native session IDs for hooks and transcript usage accounting, then close and remove their owned native transcript after use.
  • Improved first-run Claude setup handling and preserved startup progress and actionable errors in the CLI view. If further setup or sign-in is needed, run claude once in Minnow Terminal, complete its prompts, then retry.
  • Improved interactive tool handoffs so tools can run before Claude commits its final response. Late text, reasoning summaries, and usage survive generation boundaries and are attributed once; repeated native tool calls are matched without duplicating results.
  • Added transcript-based reasoning-summary collection and safer warm follow-ups and recovery. Summary timing follows Claude’s native transcript availability; completed native responses supply token counts, and unavailable cost or quota measurements stay unavailable.
  • Interactive Claude does not support the legacy CLI dollar-budget flag. A configured cap now produces an actionable request to clear it before interactive inference instead of being silently ignored.
  • Fixed Codex reasoning display to retain summary and content deltas, recover text from completed items when needed, and separate sections without repeating streamed output.
  • The Codex context ring now prefers native context occupancy and the native window, retaining them through tool handoffs and reloads. Provider/model or history mismatches, clearing, and compaction invalidate that measurement; billing totals remain separate.

Sub-agents, search, and Brain

  • Redesigned sub-agent cards as single timeline steps, replacing duplicate spawn rows. Each step shows the agent, brief, status, tool-call count, and completed-run duration, with one detail line for live activity, an error, or a result headline.
  • Kept parallel spawn cards visible after their tool batch in spawn order, instead of hiding them inside a collapsed batch. Removed the continuous status pulse to reduce rendering overhead during long runs.
  • Fixed literal code searches containing punctuation, backslashes, or leading dashes. Fixed-string searches now preserve the original snippet, and queries beginning with a dash are treated as patterns rather than command flags. Regular expression search remains available.
  • Fixed Brain graph truncation dropping every tag when the node limit was exceeded. Limited graphs retain pages alongside their tags and surviving edges, with the hidden count based on the actual removed nodes.

Design reference and maintenance

  • Replaced the placeholder frontend design reference with comprehensive guidance covering creative direction, layout, typography, color, components, UX copy, responsive behavior, accessibility, motion, performance, and visual verification.
  • Included the reference in desktop packages, added package-inclusion and content checks, and corrected recovery advice for a missing reference asset.
  • Removed the redundant server-count header from the development server screen.
  • Removed stale notes, story and todo artifacts, unsupported QA probes, an old composer snapshot, a CI log, and stray generated TypeScript output. Added ignore patterns and archived the completed repository cleanup plan.
  • Added GitHub banner artwork and its regeneration prompt, and updated the shipped manual, technical reference, product wiki entries, regression tests, and local Claude streaming and recovery smoke fixtures.

Downloads and updates

Packages are included for Windows, Linux x86_64, and Apple Silicon macOS. Installed users can check for published releases under Settings → About & troubleshooting → Updates.

Full commit comparison

v0.1.10

Minnow 0.1.10

Minnow 0.1.10 adds plugin apps and UI extensions, account usage for Codex and Claude, and more reliable CLI conversations. It also makes board failures easier to investigate, brings missing GitHub issues into Sync all, preserves composer drafts while browsing prompt history, and fixes update restart hangs.

Highlights

  • Extend the workspace with plugins. Plugins can now add apps, menus, command palette actions, direct slash commands, and UI beside chat metrics. Install or enable an extension and its UI loads without restarting Minnow.
  • See your account allowance. Codex and Claude subscription usage appears in Models → CLIs and beside the chat model, with remaining percentages, reset times, and refresh controls.
  • Keep Claude conversations intact. Screenshot results stay in the current native conversation, and verified tool-handoff checkpoints improve recovery after interruptions.
  • Understand failed board runs. Integration failures show the recorded check, exit code, command output, and expandable evidence directly in the report.
  • Recover your unsent draft. Prompt history restores your draft, selection, and scroll position; Alt+Up/Down explicitly browses history from long or multiline text.

All changes since 0.1.9

Plugins

  • Added trusted frontend modules to API v1 plugin packages, including UI-only packages. Extensions load at startup and reload when packages are installed, updated, reloaded, enabled, disabled, or removed.
  • Added plugin apps to the app rail, app command palette, and hash routing, with namespaced identities, remembered app preferences, and cleanup when an extension unloads. Plugin apps open in the main window.
  • Added menu contributions, a Plugin commands menubar menu, command palette actions, and custom menus using Minnow’s shared menu renderer.
  • Added direct plugin slash commands to the composer picker, with namespaced commands and optional aliases that cannot replace existing commands or skills. These actions can run while a reply is streaming and do not require model generation; command errors preserve the input for correction.
  • Added managed DOM mounting and stable slots for the menubar, throughput indicators, and message metrics. Managed mounts follow replaced UI elements and dispose their subscriptions when removed.
  • Added chat and workspace usage APIs for plugin token counters, including completed-request totals, cost, and live or last metrics. Usage snapshots exclude message text and credentials and remain scoped to the current workspace.
  • Added a chat token counter example plugin with a counter beside throughput, a usage app, and menu and command palette actions. Updated the plugin manual and /build-plugin authoring guidance.
  • Settings now identifies trusted UI entries and displays UI load errors. Installation explains that frontend modules run in the host document with DOM and authenticated API access; failed activation rolls back managed contributions.

Account usage and agent CLIs

  • Added Codex and Claude subscription allowance displays in Models → CLIs and a Usage button beside the Code chat model. The button shows the remaining percentage for the most-used quota window.
  • Added quota-window details, reset times, last-refresh information, automatic refresh while visible, and manual refresh with provider retry handling. Failed refreshes mark retained measurements as last-known usage; missing limits show as unavailable.
  • Account usage includes activity outside Minnow and is separate from chat token totals. Checks do not generate a model response. Codex requires a ChatGPT login and Claude requires a subscription OAuth login; API configurations do not report subscription quota.
  • Enabled Minnow tool calls for Codex models that require code mode. Tool descriptions retain the original Minnow tool names, and newer Codex native environment and delegation features stay disabled so calls use Minnow’s shared tool runner.
  • Preserved Claude’s native session when screenshot images arrive through tool results, avoiding a process restart and replay of the growing conversation.
  • Added verified Claude checkpoints at tool handoffs. After an interruption, matching saved conversations can resume once all pending tool results are recorded; mismatched history or configuration rebuilds from Minnow history without executing recorded tools again.
  • Kept CLI tool catalogs stable throughout a turn by supplying the full permitted catalog from the outset. Clarified Claude’s advertised MCP tool names in its instructions.
  • Preserved Claude text, reasoning, and final usage that arrive after the first tool call. Added native rate-limit observations, resolved model aliases, and recovery reasons to CLI diagnostics, with complete usage across continuations.
  • Fixed CLI output visibility when switching to alternate Code views, including the Brain map, and restored it when returning to chat.

Issues and workspaces

  • Sync all now discovers missing open and closed GitHub issues before pushing unlinked cards and syncing linked issues. Imports go into Triage, retain remote metadata, avoid duplicate imports and redundant pushes, and run before parent-first synchronization.
  • Discovery follows the selected workspace scope. All workspaces includes the current workspace and other workspaces represented by local issues; automatic linked-only sync keeps its existing scope. Discovery uses the existing limit of 500 GitHub issues per workspace.
  • Improved issue detail layout with wrapping editable titles, labelled Type/Status/Priority controls, and a tidier labels and workflow footer. Fixed empty description placeholders overlapping focused editing.
  • Added hierarchy connector lines for nested child issues in the issue list.
  • Run-target controls now show the current workspace by name in both the chat composer and Issues → Send to chat.

Orchestrate boards

  • Integration failure reports now show a meaningful headline from the recorded evidence, including the failed check and exit code when available. Command output appears in a bounded, keyboard-scrollable disclosure, with remaining structured evidence under Check details.
  • Attempt blocker summaries now keep every blocker expandable instead of truncating the list. Restored collapsible task rows and aligned Runs and Files in a consistent two-column layout.
  • Task-detail conversations always use the full transcript view. Activity stays in the reading flow, thoughts start expanded, tool results start closed, and individual disclosure choices survive live updates.
  • Fixed board transcript tool rendering to retain code changes and attachments, and to avoid showing an empty result for a tool that is still pending. Patch tool summaries now show the affected file or file count and whether the patch was applied.
  • Preserved round usage, runtime, and CLI diagnostics in attempt transcripts. Rejected builder and tester reports now return the complete required argument schema in the first error, reducing retries for missing fields while retaining validation.

Composer and recovery

  • Prompt history now preserves unsent text, caret selection, and scroll position. Returning from history restores the draft, and edits to recalled entries survive navigation between entries.
  • Browsing history no longer overwrites the unsent draft on blur, chat switching, or session save. Pressing Down at the draft tail leaves it intact.
  • Plain Up/Down recalls history only at the edges of a single-line composer. Added Alt+Up/Down for explicit recall from anywhere, including wrapped and multiline prompts.
  • Restored composer scrolling after clears, exposed native scrollbars, and aligned skill highlighting with the visible text width.
  • Continue on a failed turn now clears stale failure banners, markers, and recovery controls while retaining partial output and tool results. A subsequent failure displays its own notice.

Startup, updates, and maintenance

  • Added startup recovery controls that work even when the application module fails to load. Failed startup offers Reload Minnow, and stalled startup offers retry controls without prematurely marking the workspace ready.
  • Kept the paired-device connection screen visible during startup and preserved the saved device credential when retrying a connection.
  • Fixed update restart hangs caused by upgraded WebSocket connections keeping the packaged server open. Shutdown now closes those connections before installing the downloaded update; Windows updates apply silently and relaunch Minnow.
  • Added optional native-call crash breadcrumbs through MINNOW_NATIVE_CRASH_TRACE=1, recording operation status and caller stacks without argument or result contents.
  • Expanded regression coverage and local smoke tools for account usage, plugin UI, startup recovery, update shutdown, Codex tool calls, Claude screenshot and interruption recovery, board transcripts and reports, GitHub imports, and composer history. Removed the unused issue hierarchy QA fixture.
  • Added a detailed context-compression guide and updated the shipped manual, technical reference, product wiki entries, keyboard shortcuts, and archived 0.1.9 release notes.

Downloads and updates

Packages are included for Windows, Linux x86_64, and Apple Silicon macOS. Installed users can check for published releases under Settings → About & troubleshooting → Updates.

Full commit comparison

v0.1.9

Minnow 0.1.9

Minnow 0.1.9 makes day-to-day development easier to navigate, with a visual Git history, a richer pull request browser, simpler Settings, and conversations that continue across native CLI turns. It also improves mobile access, model configuration, notifications, and recovery when work is interrupted.

Highlights

  • Explore Git history visually. Source Control now has a spatial commit map with pan, zoom, a minimap, search, filters, and collapsible history. Switch to a chronological list anytime, and review a selected commit with file tabs and side-by-side diffs.
  • Review pull requests in one place. Browse PRs by state, search the loaded list, and inspect Overview, Checks, Files, and Commits without leaving Minnow.
  • Spend less time configuring. Settings has clearer navigation, focused pages, and advanced options tucked away until needed. Settings and Models panels save changes automatically.
  • Keep CLI conversations going. Claude Code, Cursor Agent, and Codex retain eligible native conversation state between turns, with recovery from Minnow history when a session cannot be resumed.
  • Use the same workspace from your phone. Paired devices use the full workspace shell and shared composer, with improved startup, reconnection, and layouts for smaller screens.

All changes since 0.1.8

Source Control

  • Added the Git history map and a remembered Map/List switch. Both views share commit selection, search, filters, collapsed groups, and review actions.
  • Added paged history across refs, with branch and remote labels. Loading older commits and refreshing history preserve your position, zoom, and focus; large histories render only the visible scene.
  • Added commit review beneath history, with horizontal file tabs and the shared side-by-side diff viewer.
  • Redesigned the pull request browser with Open, Merged, Closed, and All filters, search, and current-branch selection.
  • Added the PR Overview/Checks/Files/Commits inspector, rendered descriptions, CI links, and review and merge summaries. Refreshes preserve the selected tab, scroll position, and an open new-PR form.
  • Consolidated merge-method selection, omitted merge controls for draft PRs, and disabled direct merging for conflicting PRs.

Settings and setup

  • Simplified Settings navigation and reorganized it into six expandable groups: General, Code & workspace, AI & agents, Extensions, Data & privacy, and About & troubleshooting. Each destination opens its own page.
  • Moved less-used controls into advanced disclosures. Settings search opens the requested field and its disclosure, and existing section and field links still resolve.
  • Separated Skills and Skills Library into their own pages, refreshed agent-pack management, and made settings forms adapt more consistently to narrow, split, and floating panes.
  • Redesigned provider connections as cards with provider logos, advanced options, and a Test connection action that checks the saved endpoint without generating a reply.
  • Replaced manual Save actions across Settings and Models configuration panels with automatic saving. Rapid edits are combined, and returned settings values are merged without overwriting inputs you are still editing.
  • Refreshed first-run setup, including provider connection and default-model selection, and added desktop Interface zoom to the appearance step. Theme and zoom changes apply immediately; zoom is also available in Settings → Appearance.
  • Added notification preference resets and a filter for modified preferences.

Models and agent CLIs

  • Added recommended sampler presets for supported DeepSeek R1, Qwen3, Qwen3.5, Qwen3.6, Qwen3.8, Qwen3 Coder Next, and Gemma 4 models. Presets update documented sampling values while preserving output limits and other overrides.
  • Added Unified KV cache to the Models Load tab. Share the selected context capacity across parallel slots, or allocate the full context per slot; the memory estimate follows the selection.
  • Redesigned Models → CLIs as compact expandable rows with installation and authentication status, a provider switch, inline advanced controls, automatic saving, and validation and retry handling.
  • Added persistent native conversations for Claude Code, Cursor ACP, and Codex app-server. Each user turn starts a fresh process and resumes eligible saved state; interrupted or mismatched conversations rebuild from Minnow history. Claude’s per-process budget resets between turns.
  • Improved native CLI lifecycle management, cancellation, cleanup, and session, stream, and usage metadata. Fixed turn-context preservation for seeded conversations with no text input.
  • Grouped Cursor effort and Fast variants by model family in the picker. Added per-chat reasoning and Fast controls that select supported advertised variants, including a compact-layout setting.
  • Made the reasoning-effort footer menu easier to recognize and restored its native picker behavior.
  • Added an explicit inherited-model choice and reset actions for agent model routing, so overrides can return to the current chat model.
  • Moved the expanded CLI view’s Chat/CLI toggle into its header for easier access.

Mobile, chat, and accessibility

  • Kept paired LAN devices on the full workspace shell and shared composer. The connection screen now waits for authorization and workspace chrome, and connection checks retry after suspension, wake, and network restoration.
  • Fixed browser-persisted ID creation on LAN connections for onboarding, issue drafts, model pools, user rules, and plugin channels.
  • Improved responsive chat layouts, narrow-pane controls, and touch targets. Mobile viewport handling follows the software keyboard, and composer popovers reposition when the visible viewport changes.
  • Improved keyboard focus in mobile drawers and composer popovers, control labels, accessible status announcements, and Markdown link handling.
  • Preserved chat scrolling when the composer or surrounding chat panels resize.

Reliability and performance

  • Fixed false save conflicts between chat windows by saving only changed chats after startup and limiting startup repairs to the affected records.
  • Routed desktop alerts through Electron’s native notification bridge, with click navigation and a notification test action in Settings.
  • Improved bulk GitHub issue synchronization with three concurrent workers. Parent issues finish before sub-issues begin, while per-issue locking and safe storage updates remain in place.
  • Fixed recovery for stuck Orchestrate boards and cancellation of stale writes.
  • Fixed Brain’s background code indexer changing the saved workspace when indexing a chat or board worktree. Restarting Minnow now retains the user’s project selection.

Downloads and updates

Packages are included for Windows, Linux x86_64, and Apple Silicon macOS. Once this release is published, installed users can check for it under Settings → About & troubleshooting → Updates.

Full commit comparison

v0.1.8

Minnow 0.1.8

Minnow 0.1.8 improves everyday work in Code: editing a message keeps its attached files, quick issue capture uses the full issue form, and reloading waits for chats to save. This release also improves companion reconnection, preview overlays, board reasoning settings, and desktop stability.

Highlights

  • Edit messages without losing files — Attached file snapshots return to the composer as removable attachments when you edit a message, so resending keeps the file context you selected.
  • Full issue capture, wherever you work — Quick capture opens the full New issue form without leaving your current app. Expand and Create saves the issue immediately and expands it in the background.
  • Safer reloads — F5 and Ctrl/Cmd+R wait for an acknowledged chat save. If saving fails, the reload is cancelled so unsaved work stays available.
  • Companion reconnects automatically — Pairing survives workspace changes, page reloads, Minnow restarts, and temporary network outages.

Chat and session saving

  • Editing a user message restores its attached file snapshots as removable pending attachments and keeps raw file markup out of the editable prompt. Resends and forked history retain the selected file context.
  • Long tool-command chips truncate within the transcript instead of overflowing their row.
  • A chat changed in another window no longer blocks unrelated edits or new chats from saving. The conflicted chat keeps its unsaved changes and shows a save warning.
  • Manual reload waits for chat saving to complete and marks active turns as interrupted for recovery. A failed save cancels the reload.

Issues

  • Quick capture reuses the full New issue form, including description, type, priority, and labels, while keeping the current app in the foreground.
  • Captured code, Git, issue, and chat references remain linked to the created issue in the current workspace.
  • Expand and Create closes the form after creating the issue and runs AI expansion in the background. Expansion preserves fields you edit while it runs; if expansion fails, the created issue remains available.
  • Narrow issue lists hide secondary project, agent, rollup, and count columns to give titles more room, including when the detail pane is open.

Code and Source Control

  • Fixed Code map indexing in installed builds when Minnow starts from a folder outside its installation: the index worker now finds the bundled language-server resources correctly.
  • Deleting an unmerged local branch now offers an explicit force-delete confirmation. This works for single and bulk deletion, with other deletion errors reported separately.

Browser previews

  • Command palettes, New issue, context menus and submenus, chat search, Git name entry, keyboard help, and model-fork dialogs stay above native browser previews, including split panes.
  • Opening a shell popover retains the preview’s last painted frame while the live browser is temporarily hidden, avoiding a blank preview beneath the menu.
  • Resizing or binding a preview while an overlay is open no longer brings the browser back above it; live previews return after the last overlay closes.

Orchestrate boards

  • Board reasoning settings correctly save and apply every supported effort level, including minimal, xhigh, and max.
  • New worker attempts enable thinking for an explicitly selected effort and retain that level in their transcript and model request. Active attempts keep their current setting.

Companion

  • Saved pairing survives workspace changes, page reloads, and host restarts. After a Wi-Fi interruption or host outage, the companion reconnects automatically.
  • Reopening a used QR link in an already paired browser keeps the existing pairing. A new code is needed only if the device is revoked or its saved browser credential is removed; use the same browser and host address to retain that credential.

Desktop and agent CLIs

  • Default interface zoom is now 100%, up from 80%. Existing saved zoom preferences are retained.
  • Added a mitigation for an Electron main-process crash traced to V8’s fast JSON serializer. Minnow uses the standard serializer in the main process to avoid that crash path.
  • On Windows, Codex launches its native executable when available, keeping background runs hidden. Custom installs can still fall back to the Node launcher.

Downloads

  • Windows: Minnow-Setup-0.1.8.exe
  • Linux: Minnow-0.1.8-x86_64.AppImage (make it executable with chmod +x before running)
  • macOS: Minnow-0.1.8-arm64.dmg

Full Changelog: https://github.com/HenriGrimm/Minnow/compare/v0.1.7…v0.1.8

v0.1.7

Minnow 0.1.7

This is a big one! On our way to a 1.0 release, Minnow 0.1.7 redesigns the Code chat, adds plugins, backup / restore functions, a rebuilt Code map, and makes Orchestrate boards editable while they run. It also carries a large batch of security, persistence, and reliability fixes.

Highlights

  • Redesigned Code chat — A cleaner transcript where finished turns fold into one summary line and live turns show a timeline of steps, plus a floating composer with mode, branch, reasoning, context and model on a single row.
  • Plugins — Install and manage plugin packages from Settings → Plugins. A plugin can add tools, encrypted connection fields, sandboxed panels and skills, and /build-plugin helps an agent author one.
  • Backup and restore — Save chats, Brain, settings and credentials to a single file, take scheduled snapshots, and restore on this or another computer from Settings → General. Also available as minnow backup and minnow restore.
  • New Code map — A pannable map of your code at three depths (Architecture, Files, Call graph) with search, an inspector, and a chat sidebar for asking about what you select.
  • Editable boards — Edit task specs, re-sync a board from its plan, skip a task by hand, and choose the starting and landing branches.
  • Scheduler works in installed builds — Scheduled jobs now run in the packaged app; previously they only ran from source.

Chat

  • Redesigned transcript and composer: prompt blocks for your turns, a single work line for settled turns, a step timeline for live ones, a run-status line above the composer, and a new-chat home with recent threads.
  • /followup hands work on to a new chat seeded with a summary of the current one. Chain up to 10 follow-ups, with your own prompts or agent-chosen tasks.
  • Copy a whole transcript, or export it as a shareable HTML file.
  • Tool calls made in the same model round collapse into one batch row with per-tool counts.
  • New wait tool lets an agent pause on a timer (up to 2 hours) instead of polling, with a visible waiting state and a notification.
  • New apply_patch tool for multi-file edits with rollback, tolerant of minor patch formatting slips.
  • Queued messages stay in the composer until they are sent, and queued commands such as /compact run as commands instead of being sent as text.
  • Thoughts panels stay readable in compact view, and Responses-API models now return reasoning summaries.
  • Token speed is measured as real output over real time, and updates live during reasoning.
  • Faster first turn: messaging is warmed while you type and the prompt appears before worktree setup finishes.
  • Tidier “calling” and “running” tool indicators, and the composer shadow no longer bleeds over the bottom bar.
  • Fixed: mode switching after a chat completes, a stale “generating” status after streams end, pending attachments leaking between chats, and requests rejected by strict OpenAI-compatible servers when a chat contained an image (thanks @Rhonstin).

Context and models

  • The context budget now uses the model’s full context window by default. An optional cap is available under Settings → Agents → Context policy.
  • Default maximum output raised from 32k to 131k tokens, and work agents continue when a round ends at the output limit.
  • The 16 MiB limit on a single streamed generation is removed.
  • Compaction keeps bounded source excerpts and labelled findings, so agents recover file context after a compaction.
  • Reasoning levels come from the provider’s catalog when it advertises them, including DeepSeek V4 efforts. The model menu has a clearer default-reasoning control, saved per model.
  • Context7 docs lookup works without an API key, and web and plan-edit tools are enabled by default.
  • Brain memory auto-injection is stricter, so broad phrasing no longer pulls in unrelated pages.
  • Fixed the Ollama model probe requesting /v1/v1/models, and corrected the documented Ollama base URL.

Agent CLIs (Claude Code, Codex, Cursor)

  • CLI processes stay alive across a turn’s tool rounds instead of restarting and replaying the transcript.
  • Codex follow-ups reuse a retained app-server conversation (Codex CLI 0.153.4 and later).
  • Live CLI output is shown in the chat.
  • Configurable context windows per agent CLI, and CLI providers now appear in Settings → Providers.
  • Codex models are discovered through the installed CLI, Codex reasoning summaries are shown, and Claude CLI aliases show versioned labels.
  • CLI login status is checked when Models opens.
  • Fixed excessive CLI usage caused by logs and tool retries.

Orchestrate boards

  • Edit a task’s title, build, test, accept and touches from the board, and re-sync a board from its plan with a dry-run preview.
  • Skip a task by hand to unblock its dependents, and merge or skip work retained from a task.
  • Choose the board’s starting branch and where it lands: the original branch, an existing branch or a new one.
  • Plan dependencies are validated when a board is created, with line-numbered errors, and a new plan check tool validates a plan beforehand.
  • Plan repair runs as a normal Plan chat turn in the sidebar.
  • Task detail has a toolbar with the card’s actions and opens on the live agent thread.
  • Board-wide live token metrics, preserved task durations, and tasks that stay fully expanded.
  • Reasoning level can be changed on a running board, and attempts with no level picked follow Settings → Thinking.
  • Interrupted attempts resume in place without spending retry budget.
  • Abandoning a task now stops its agents and provider streams.
  • Autopilot has a per-attempt time limit (240 minutes by default, configurable or off). Sub-agent timeouts are removed, and testers have round limits.
  • Work already in the pipeline is scheduled ahead of new builders, and run reports are more compact.
  • Creating a board from a plan that already has one opens the existing board.
  • pnpm workspaces get a real install per worktree, and an agent install can no longer rewrite the main checkout’s dependencies.
  • Browser automation is no longer available to board agents.
  • Open transcript details stay open while streaming, and completed thoughts stay visible.

Code

  • Code map rebuilt as a map-first page with Architecture, Files and Call graph depths, a floating inspector, Ctrl+K search, and Mermaid and PNG export.
  • Ask about a selection in a resizable chat sidebar beside the map; the question sends straight away with a context card.
  • Selecting a file in a commit diff and on the map stays in sync.
  • Sidebar search is backed by a cached workspace file catalog, so filename matches appear immediately, and the file tree can search file contents.
  • Godot support: GDScript, resource and shader editing, managed Godot 4 discovery and install, and agent tools to inspect and control a Godot project.
  • The editor reconciles external edits before saving and keeps newer drafts across pending saves.
  • Dev server detection runs as a normal Build chat, and the Dev Servers screen is tidier.
  • In-app browser gains an actions menu with secondary toolbar actions consolidated into it.
  • Deferred onboarding steps in Code are now actionable.

Source Control

  • Filterable branch switcher with Local and Remote groups, worktree-locked branches, and create-from-HEAD.
  • Open a commit’s diff from history, including without an open file.
  • Minnow asks before publishing a new remote branch.
  • Fixed row selection, bulk-delete progress, the branch picker after removing a worktree, and duplicate board worktree names in the picker.

Issues

  • A View selector replaces the saved-filter tabs.
  • Copy issue: a full Markdown copy of an issue from the peek actions menu.
  • Categorisation and comments sync through GitHub, with pending-sync badges in list, board and detail views.
  • Signal-bar priority icons, with a warning triangle for urgent.
  • Formatted tickets and images render in chat, linked chats open from an issue, and nested lists and HTML source can be edited.
  • Label suggestions are scoped to the destination workspace, and the rail badge counts the active workspace only.
  • Fixed: duplicated GitHub metadata text in descriptions, remote issues lost after a failed first load, new-issue descriptions and labels lost on reopen, issue IDs colliding across workspaces, and a hidden Backlog lane on older catalogs.

App and navigation

  • Themed dropdown menus across the app.
  • Command palette has more actions and fewer redundant entries.
  • Global search shows recent chats and has category filters.
  • Smoother window resizing and less background rendering work.
  • Faster startup.
  • MCP hub: persistent scoped credentials, clearer stdio setup, and the workspace is resolved from the requesting agent.
  • Brain graph shows pages and tags only; code symbols now live in the Code map.
  • Deep Research settings are removed.
  • Fixed: macOS header spacing in full screen, switch thumb alignment, and voice startup after a socket failure.

Security

  • Workspace previews are served from a separate, restricted origin, and the host session token no longer appears in preview URLs.
  • Host token injection requires a loopback peer.
  • Headless API requests are authenticated.
  • Hardened webhook delivery, and guarded Electron shell IPC and navigation.

Reliability

  • Scheduler: jobs run in installed builds, interrupted jobs recover on startup, due jobs fill free slots concurrently, a failed preparation no longer leaks a run slot, and child output is bounded.
  • Sessions: concurrent chat writers are guarded by revisions.
  • Brain: page saves are atomic and revision-aware, quoted frontmatter round-trips, and vector index updates are serialised.
  • Skills: installs are staged and keep local changes.
  • Sub-agent results are held until the parent accepts them.
  • MCP stdio servers launch correctly from the installed app.
  • LAN access is honoured in the installed app.
  • Headless CLI: runs fail on generation errors, cancel on interruption, and keep the caller’s working directory and workspace.
  • Server streams close cleanly on an update restart, and disconnected Git and LSP requests cancel their own work.
  • Attachment extraction is bounded and cancellable.

Downloads

  • Windows: Minnow-Setup-0.1.7.exe
  • Linux: Minnow-0.1.7-x86_64.AppImage (make it executable with chmod +x before running)
  • macOS: Minnow-0.1.7-arm64.dmg

New contributors

Full changes: https://github.com/HenriGrimm/Minnow/compare/v0.1.6…v0.1.7

v0.1.7-beta.20261001.21.1 PREVIEW

Minnow 0.1.7-beta.20261001.21.1 (pre-release)

Automated beta build from commit 6f814ee0fdedb037928fec415b3ed566d91841a5.

What’s Changed

Full Changelog: https://github.com/HenriGrimm/Minnow/compare/v0.1.7-beta.20260930.20.1…v0.1.7-beta.20261001.21.1

v0.1.6

Minnow 0.1.6

Minnow 0.1.6 adds an MCP hub for connecting external agents and a clearer view of token usage.

Highlights

  • MCP hub — Connect an external agent to Minnow’s Issues, Brain, and user manual through HTTP or a local command. Choose read-only or read-and-write access and copy the configuration from Settings → Integrations.
  • Usage timeline — Explore token use over time in Settings, with time-range controls and a breakdown of prompt and completion tokens.
  • Reliability — MCP discovery handles failed connections more gracefully, and Orchestrate restores a builder’s worktree snapshot without merging newer integration changes into it.

Downloads

  • Windows: Minnow-Setup-0.1.6.exe
  • Linux: Minnow-0.1.6-x86_64.AppImage (make it executable with chmod +x before running)
  • macOS: Minnow-0.1.6-arm64.dmg

Full changes: https://github.com/HenriGrimm/Minnow/compare/v0.1.5…v0.1.6

v0.1.5

Minnow 0.1.5

Minnow 0.1.5 improves the day-to-day build workflow with a richer Issues view, more capable source control, and better visibility into agent work.

Highlights

  • Issues projects — Browse projects and move between issues with improved navigation. Issue mentions in chat now link to their issue, and issue views render more reliably.
  • Source control — Explore branches as a tree, select multiple branches, and see an inferred parent branch when working with branch trees.
  • Agent CLI — Improved installation and discovery, Windows PATH resolution, and live reasoning and tool activity in chat.
  • Preview — Browser history and address-bar suggestions make it easier to revisit pages while building.
  • Models — Model configuration now lives in the Models app. A per-load vision toggle lets you load a model without its vision projector when needed.
  • Workspace polish — Collapse the right pane without closing its tabs, drag files from the tree into native apps, and receive desktop notifications with configurable settings.
  • Reliability — Fixes for runner retries, output-token estimates, issue rendering, and board tester recovery.

Downloads

  • Windows: Minnow-Setup-0.1.5.exe
  • Linux: Minnow-0.1.5-x86_64.AppImage (make it executable with chmod +x before running)
  • macOS: Waiting for Notorization

This release remains a draft while the macOS package is prepared. Draft releases do not appear in the in-app updater.

Full changes: https://github.com/HenriGrimm/Minnow/compare/v0.1.4…v0.1.5

v0.1.4

Minnow 0.1.4

Eighth open-beta build. Voice dictation works out of the box, context compaction v2 keeps long chats under control, and a new Home app tracks AI activity across your workspace — plus a fix for the v0.1.3 packaged-app boot failure.

Highlights

  • Context compaction v2 — Server-side context policy, per-turn/round compaction, compact + recall_history, and better local reasoning replay with accurate token estimation.
  • Home app — New rail app with AI code activity tracking across your workspace.
  • Models Discover redesign — Cleaner Discover and download flow, with creator avatars on model cards.
  • Voice dictation — Improved Built-in dictation with system speech defaults; no more extra setup.
  • Preview — Design strip now mirrors with CDP hover picking: point at a live preview to pick the element to restyle.
  • Packaged boot fix — v0.1.3 packaged builds could fail to start with a misleading fetch failed; v0.1.4 resolves it.
  • MCP setup simplified — Fewer steps to connect an MCP server, with a simpler OAuth flow.

Changes and fixes since v0.1.3

Voice

  • Built-in dictation and system speech defaults.

Context and chat

  • Context compaction v2: server-side context policy, per-turn and per-round compaction, compact command and recall_history.
  • Local reasoning replay improvements and more accurate context token estimation (including llama.cpp timings).
  • Adaptive thinking display for Anthropic models.
  • Per-turn summary tally and activity line in the chat work view; turn todo checklist moved from the composer into the transcript.
  • Chat scroll stays put when the transcript updates late.

Home and models

  • New Home app with AI code activity tracking.
  • Redesigned Models Discover and download flow; creator avatar images on model cards.

Preview

  • Design strip mirror with CDP hover picking.
  • Design strip icons embedded as inline SVG in the native guest (fixes strip icons not rendering in the packaged app).

Orchestrator

  • Generated commits use board task titles; merge-queue re-sends tracked; plan paths in task seeds.

Brain

  • Tags are the default view on the knowledge graph.

Issues

  • Quick capture simplified to a minimal popover; commit generation status inline.

Electron and packaging

  • Fixed the v0.1.3 packaged boot failure (fetch failed — harness registry now ships in app.asar) and the workspace claim transport issue.
  • Windows branding differentiates packaged vs development builds.

Under the hood

  • MCP server setup and OAuth simplified.
  • GitHub CLI integration refactored onto a shared runGh helper; worktree cleanup inspects husk state.
  • Browser navigation and app window allowlists enhanced.
v0.1.3

Minnow 0.1.3

Seventh open-beta build. Source Control is now a first-class app, chat transcripts can switch between Compact and Full views, and context usage is easier to read.

Highlights

  • Source Control app — Dedicated rail app for changes, history, branches, worktrees, pull requests, and CI. Opens without clearing your Code chat.
  • Compact and Full chat views (#1194) — Compact groups per-turn Working/Worked activity and shows edited-file summaries with inline diff review; Full keeps every transcript row visible.
  • Context usage — Improved usage ring, stats strip, and web content extraction for clearer context budgeting.
  • Default model configuration — New chats and the menubar model chip resolve the configured default provider/model consistently.
  • Issues — Clearer closed-issue handling and list/board indicators.
  • Orchestrate — Board baseline now returns seq for reliable reconnect sync.

Changes and fixes since v0.1.2

Source Control

  • Released core app at #/app/source-control with the full git surface (changes, history, branches, stashes, worktrees, PRs, CI).
  • Code left-pane Source Control toggle still works; the rail app opens the same capabilities in a dedicated view.

Chat and context

  • Settings → Appearance → Chat view: Compact (default) or Full, persisted across restarts.
  • Per-turn file-change summaries and inline diff review in Compact mode.
  • Token/stat formatting handles trillion-scale counts.

Models and settings

  • Default model binding management across Settings, menubar chip, and new chats.
  • Full-width skill preview buttons; Impeccable skill sync updates.
v0.1.2

Minnow 0.1.2

What’s Changed

A broad update across model routing, browser automation, Issues, Orchestrate, chat reliability, and Git workflows. This release also improves tool loading and streaming, workspace handling, and the desktop build pipeline.

Models & providers

  • Model pools — share capacity across models and providers, with automatic failover and clearer quota-exhaustion notifications.
  • Multi-GPU GGUF loading — split GGUF models across selected GPUs from the Models inspector.
  • OpenCode Go support — route requests with the correct provider identity and headers.
  • Improved OpenAI-compatible response handling, provider discovery, and utility-model routing.

Agent Browser & tools

  • Agent Browser — added browser automation with an integrated viewer, runtime controls, watchdog handling, and expanded test coverage.
  • Lazy tool schemas — tool definitions can load on demand, reducing initial prompt weight while preserving list-only discovery.
  • Added configurable context injection settings and improved tool streaming, output limits, slicing, caching, and transcript presentation.
  • Improved Impeccable integration and built-in skill chips.

Issues

  • Custom issue types now support colors and icons.
  • Expanded label management, attachment handling, issue expansion, and run-target selection.
  • Improved GitHub two-way synchronization, workspace handling, deletion behavior, and error reporting.
  • Issue detail now includes linked chats and boards plus a collapsible files panel.

Orchestrate

  • Added Reset and Rewind board actions.
  • Added richer end-of-run reports, task evidence, notes, and file presentation.
  • Hardened board recovery, merge queues, rebase seeds, worktree cleanup, and operational merge diagnostics.
  • Improved task detail and attempt transcript handling.

Chat & context

  • Improved streaming paint behavior, scrolling, thought persistence, and transcript rendering.
  • Stopped turns keep their context when continued.
  • Refined context trimming, screenshot token pricing, usage metrics, and the context ring.
  • Improved per-chat question handling, mode refresh behavior, message actions, and queued activity.

Git, workspaces & desktop

  • Improved Git feedback, operation errors, branch/worktree creation, and merge-failure diagnostics.
  • Gitmoji shortcodes expand to Unicode in commit messages.
  • Released apps can open in separate windows.
  • Improved Electron workspace management, launch handling, and packaging validation.

UI, docs & maintenance

  • Improved Markdown link handling and previews.
  • Refined Settings layouts, model-pool controls, injection settings, and related app styling.
  • Made generated skill metadata deterministic and updated built-in tool definitions, tests, documentation, and product-wiki metadata.
v0.1.1

Minnow 0.1.1

What’s Changed

A big update for the Issues tracker — label catalog, sub-issues, a redesigned issue detail view, and a much deeper GitHub two-way sync — plus Git Bash terminal support on Windows and server-backed appearance persistence.

Issues

  • Label catalog — named labels with color swatches, right-click to recolor, up to three chips plus an overflow popover, and a + add flyout in list and board views.
  • Sub-issues — one level of nesting: add/unparent children from the peek, or drop an issue onto another issue to make it a child.
  • Issue detail redesign — sticky identity header, property chips, and workflow; resizable peek with full-screen sheet mode; secondary sections (Code, Attachments, Git, Plan, Related) collapse to add-rows when empty.
  • Comments & activity — agent comments and activity render in issue detail.
  • Chats section — issues link the chats/boards they were created from, with Running/Done state and unlink.
  • Type & status icons — glyph picker in Settings → Issues; list and status chips render the glyphs.
  • Drag & drop refinements — issue-on-issue parent drops, board column drops, and OS file / chat-capture drops into the detail view.
  • New issue form — workspace picker (Scratch + recent workspaces) when the header scope is All workspaces.

GitHub sync

  • Sync all button in two-way mirror mode, plus automatic sync: auto-push title/description/labels/closed-state after edits, create cards on GitHub on first edit of an unlinked issue, and a background sync every 5 minutes.
  • Labels pushed to GitHub by name (missing labels are created first); dropped names surface in a toast when the repo forbids label creation.
  • Conflict handling — Keep mine / Keep GitHub prompts; sync status chips in the peek (synced / needs push / open).
  • Sync failures now show friendly local-backend copy instead of backend jargon.

Terminal

  • Git Bash support for Windows terminal profiles, plus improved shell command handling (one-shot spawns, PTY environment, sandbox signal handling).

Desktop / windows

  • Window retargeting — opening a plan or workspace reuses a window already on that folder instead of spawning a duplicate, with a guard against accidental workspace switches.

Appearance

  • Server-backed appearance persistence — theme, fonts, and custom colors persist through the tool server instead of browser-local storage alone.

Settings & Models

  • Settings → Models app navigation improvements; the Models app opens directly from settings sections.

Fixes

  • Skip the corrupt image_url capability probe on loopback llama.cpp servers, avoiding spurious probe failures on local models. (#1145)

Docs

  • v0.1.0 release notes, new plans (Git Bash terminal, issues GitHub auto-sync, sub-issues, labels, status icons, app windows), and manual updates for the Issues app and keyboard shortcuts.

Full Changelog: https://github.com/HenriGrimm/Minnow/compare/v0.1.0…v0.1.1

v0.1.0

Minnow 0.1.0

Sixth open-beta build — first 0.1.x. The board engine, session store, and workspace windows were rebuilt so the build loop holds up overnight, across restarts, and across more than one repo at a time.

Highlights

  • Orchestrator V2 — Boards are a journaled server engine, not a chat that mutates a kanban. Pick a plan, start waves of Builder/Tester agents in isolated worktrees, merge when the ladder passes. Crash, reload, and workspace switch replay from the journal. V1 board tools (board_init and siblings) are gone.
  • One folder per window — Open several workspaces at once as separate windows. Re-opening a folder focuses the window that already has it. Close asks before hiding to the tray. The workspace picker is a real welcome page.
  • Sessions that survive restart — Chat state lives in sessions.db with rotating snapshots. A lazy boot no longer wipes transcripts. Failed turns offer Continue or Clear. Opening the app asks before resuming interrupted chats and boards.
  • Shared agent runner — Product chat, boards, and sub-agents share one stream/tool loop. Sub-agent runs persist in their own journals and keep going when the parent chat is elsewhere. Context overflow summarizes and retries instead of hard-failing.
  • Chat and Code polish — Queue a follow-up while a turn is running. Drag editor and browser tabs onto chat as durable link chips. Material file-type icons in the tree and tabs. In-app Review PR from Issues and Source Control. Expand an issue into a review overlay before saving.

Changes and fixes since v0.0.5

Orchestrate

  • V2 Boards page is the only live orchestrator surface: plan picker, live kanban, task-detail overlay with chat-shaped attempt threads, follow-up chat from the finish report (#941, later V2 landings).
  • Worktree slots are named from the board, wave, and task; orphan slots reclaim in bulk; Start fails closed if git init cannot run.
  • Last-opened board resumes; running boards show an activity dot on Orchestrate; display-wake polling no longer kills live task chats (#949).
  • Broken plan markdown offers Repair (plan-repairer sub-agent) instead of a dead parse.
  • Skip the plan picker when a plan is already bound (#1027). Auto-fix illegal worktree and branch names (#958).

Workspaces and shell

  • Multi-window: one workspace per view, restored geometry, close-workspace dialog, window-close vs tray preference.
  • Welcome page / picker: Sandbox pinned, recents of folders you opened, Create project / Open folder.
  • Ask before resuming chats and boards on app open (#953).
  • Hardware acceleration setting; GPU work parks while the window is hidden or streaming.

Chat, context, and agents

  • Queued follow-ups render in the transcript (#1032). Drag editor/browser tabs onto chat as link chips (#1034).
  • Context overflow summarizes and retries; exhausted turns fail visibly with Continue / Clear (#948, #950, #959).
  • In-flight tool rows stay visible across a chat switch (#944). Interleaved think tags stay off the reply (#957).
  • Sub-agents adopt the shared runner; spawn cards re-anchor after reload; report nudges for boards and sub-agents.

Sessions and reliability

  • Canonical store is SQLite sessions.db; rotating verified snapshots replace the JSON mirror.
  • History-omitting writes cannot wipe transcripts; chat deletion is never inferred from absence.

Issues, git, and Code

  • Sparkles Expand with a review overlay — nothing saved until Apply (#1033).
  • In-app Review PR from Issues and Source Control.
  • GitHub issue import no longer freezes the shell (#961).
  • Material Icon Theme in the file tree and viewer tabs; Files / Browser / Source Control pane cluster (#1035).
  • File-tree drag-and-drop move (#939). Word wrap on git commit review (#1029).

Models, terminal, and internals

  • Vision probe on first model load (#960). MLX load and inference meters. Loading progress on Local Server.
  • llama.cpp offload-slot handling; default reasoning options for openai-v1.
  • Terminal: zsh history no longer shows ^A^K (#956); clipboard paste / copy shortcuts; Windows ancestor tracking.
  • Dev servers: do not inject Vite --port into non-Vite CLIs.
  • Appearance: Google Fonts catalog (selected pair only); wallpaper section removed (#1037).
  • Est. cost metric cells no longer overlap (#937). Configurable tool-result size limits. Landlock path-list caps.
  • Catalog is 105 built-in tools (V1 board tools removed).
v0.0.5

Minnow 0.0.5

Fifth open-beta build focused on Issues v2, local llama.cpp hosting, compact Code composer, vision and thinking models, and agent-loop reliability.

Highlights

  • Issues v2 (#892) — Capture from anywhere (menubar, Alt+C, right-click, drag-and-drop), grouping/rank/triage, markdown-preserving editor, assign-to-agent via Orchestrate, and GitHub sync with explicit conflicts.
  • Local llama.cpp (#891) — Auto-fit context/offload, crash/OOM diagnosis with retry, resumable downloads, split-GGUF shards, multi-serve LRU.
  • Compact composer (#915) — Narrow columns: mode dropdown, overflow cog, model chip, context wheel. Cog is a two-page sheet (this-turn settings → Tools).
  • Qwen3.8 + thinking (#878, #872) — Host Qwen3.8-27B; composer High → xhigh; thinking-budget cuts continue from prior reasoning, scoped per turn.
  • Agent harness (#875) — Stop mid-batch no longer crashes; unpaired tool calls are repaired; shell/MCP writes bust the file cache; rate limits retry.
  • Web search (#898) — Decoy Bing results filtered; DuckDuckGo + Stack Overflow in general; optional deep_read.
  • Brain reindex (#855) — Full reindex completes on real repos; PageRank persists; UI polls a background job.

Changes and fixes since v0.0.4

Issues

  • Issues app v2: saved views, grouping, rank, triage, sub-issues, inline editing, keyboard chrome, type icons in Settings (#892).
  • Capture: menubar, Alt+C, right-click, drag-and-drop onto rail / list / board.
  • Description editor round-trips untouched markdown blocks byte-for-byte.
  • Assign to agent via Orchestrate; GitHub sync with conflict resolution (manual).

Models and hosting

  • llama.cpp auto mode: --fit on, no -ngl, -c on a 32k-preferred ladder capped by train context (#891).
  • Serve crashes/OOM visible and retryable; downloads resume; split GGUFs; multiple resident servers.
  • Memory from attention geometry (GGUF header), not parameter count — sliding window and partial offload modelled (#854).
  • Qwen3.8-27B catalog, llama.cpp b10448, --jinja + mmproj; High maps to xhigh (#878).
  • llama.cpp /props reports live context window and vision.
  • Models runtime log scrolls inside the OS stage instead of clipping (#882).

Chat, composer, and Code

  • Compact composer + two-page overflow cog (#915).
  • Jump chip and code-change strip no longer overlap; Commit / Create PR beside undo.
  • Right pane keeps the file viewer when a preview tab stays open; preview stays painted beside stage views.
  • Streaming bubbles paint incrementally instead of only at completion (#868).
  • Drop folders from Explorer/Finder into the file tree (#886).
  • Clipboard image paste no longer duplicates; composer spellcheck and auto-resize.

Agents, thinking, and vision

  • Parallel Stop no longer throws; unpaired tool_calls repaired before send; cache invalidation for shell/MCP; 429/529 retry (#875).
  • Thinking-budget continuation carries reasoning forward; budget is per turn (#872).
  • Vision probe; Qwen/Harmony tool-call parsing; screenshots fed to vision models (#871).
  • LM Studio tok/s trusts server stats when coherent; browser_screenshot capture bounded at 3s.
  • Orchestrate boards: per-board reasoning effort and thinking mode.

Orchestrate, search, MCP, LSP

  • AFK board chats recover after stall-stop; live tool output counts as progress (#894).
  • Orchestrator git-init no longer loops on ask_question (#887).
  • Web search relevance guard; safe_search; SearXNG secret key stable across launches; deep_read (#898).
  • MCP snake_case dispatch fixed; MCP tools in Settings → Tools with persistable permissions (#881).
  • Excel/Office read_file no longer dumps ZIP bytes (#884).
  • tsserver restarts when Vite/Node project inputs change (#885).

Brain, preview, CI

  • Brain reindex: streamed usage scan, framed worker IPC, 202 + poll, failed-file summary, PageRank write (#855).
  • @emnapi lockfile restored for npm ci; issue v2 tools in capability catalog; wiki/budget refreshes.
v0.0.4

Minnow 0.0.4

Fourth open-beta build focused on Orchestrate board execution, model routing and MLX hosting, Brain code index, and memory editing.

Highlights

  • Orchestrate and boards — Synthesis config, board chat metrics, board view polish, and task worktree sync for parallel execution.
  • Model routing — Optional prompt expander binding in Settings → Model routing.
  • Models picker — Streamlined Load / Unload, host-filter resolvers, and stronger provider resolution.
  • MLX — HF hub cache created before mlx-lm spawn (#836).
  • Brain code index — Workspace-scoped indexing and API improvements.
  • Memory — Edit saved entries from the Memory UI.
  • Product wiki — Preview hides behind the manual overlay; catalog guarded in CI.
  • Chat — Clearer generation errors; improved stop-generation and agent activity handling.

Notable fixes since v0.0.3

  • Composer model menus: host filter Load / Unload wired to correct resolvers.
  • CI stall recovery; model picker CSS contract.
  • Preview visibility when wiki overlay is open; screenshot links with session token support.
  • @emnapi optional deps restored in lockfile for packaged Windows builds.
v0.0.3

Minnow 0.0.3

Third open-beta build centered on a workspace-first shell, Orchestrate library surfaces, polish across Models, Brain, and agent safety.

Highlights

  • Workspace-first boot — Workspace picker at launch; app rail and Code align to the active folder; optional fresh chat when switching workspaces (#825).
  • Slash skills — /plan-work, /orchestrate-plan (#824), and /fix-ci (#796).
  • Agent shell sandbox (opt-in) — macOS Seatbelt / Linux Landlock for one-shot agent commands; default off (#810).
  • Models — Auto-load before chat, stale-serve reconciliation, MLX context/log/shutdown fixes (#809, #823).
  • Per-workspace Issues — Project keys follow the workspace (#783).
  • Ctrl + scroll — Zoom the desktop shell (#790).

Notable fixes since v0.0.2

  • Boot dual-gate loader and workspace gate polish; boot/streaming performance (#784).
  • Enter stops chat instead of sending queued steer; Windows Open in Explorer.
  • Terminal run tracking across eviction/restart; stable grep pagination.
  • Custom GPU layers on hosting; gh CLI guidance for GitHub in agent chats.
v0.0.2

Minnow 0.0.2

Second open-beta build focused on the Code editor, Brain code map, LSP, and local models on Apple Silicon.

Highlights

  • Code map follows your workspace — Brain code index and repo map injection respect the active workspace root.
  • Packaged LSP works — Node-based language servers start correctly inside the Electron app.
  • Go (gopls) — Installs via go install with correct upstream versioning (Go on PATH).
  • Intent mode — Restored styling, idle auto-proposals, and correct accept/reindent behavior (#779).
  • MLX on Apple Silicon — Optional mlx-lm runtime and Hugging Face catalog search (#780).
  • My Models — Fixed Minnow Library provider binding (#778).

Fixes since v0.0.1

Code editor and Brain

  • Code map / Brain code index use active workspaceRoot for repo maps and injection.
  • Intent mode display, idle proposals, line-leave resolve, and accept reindent (#779).
  • Packaging no longer kills unrelated Electron apps (e.g. Cursor) during npm run package.

Language servers

  • Bundled Node LSPs spawn with Electron’s Node (ELECTRON_RUN_AS_NODE) (#768).
  • gopls via go install; Settings install/remove for Go bundles.
  • Bash LSP: ShellCheck requirement documented; improved workspace config resolution.

Models and providers

  • MLX inference and Hugging Face search on Apple Silicon (#780).
  • minnow-library provider id for hosted My Models (#778).
  • Kimi / Moonshot: omit reasoning fields on tool-loop replay when the API rejects them.

Documentation and CI

  • Docs and third-party notices aligned with current product (#775).
  • CI stability fixes (grep fixtures, capability probe, LSP completion timeouts).
v0.0.1

Minnow 0.0.1

First public preview of Minnow: a free, AGPL local-first AI workspace for chat, code, orchestration, memory on the models you already run (LM Studio, Ollama, llama.cpp, and or any OpenAI-compatible API).

This is a beta. Expect rough edges. Feedback and bug reports are welcome on GitHub Issues.

Highlights

  • Orchestrator boards — Turn a plan into a kanban line with Builder/Tester agents and isolated git worktrees.
  • 114 built-in tools (106 in default builds), sub-agents, work agents, MCP, and per-tool permissions.
  • Privacy by default — No Minnow account, no telemetry; data under ~/.minnow; loopback-only server until you opt into LAN access.